The process of adjusting weights based on error gradients is called (a)   .

The (a)   layer in an ANN receives raw input features.

Overfitting in ANN can be reduced using a technique called (a)   , where random neurons are ignored during training.

The universal function approximation theorem states that an ANN with at least one hidden layer can approximate any (a)   functions.

The (a)   algorithm is a variant of gradient descent that includes momentum and adaptive learning rates.

In an ANN, the sum of weighted inputs plus bias is passed through an (a)   to produce the output of a neuron.

A neural network without hidden layers is equivalent to performing (a)   regression.

The (a)   function is commonly used for binary classification problems in ANN.

In ANN, the weights are updated using the rule: New Weight = Old Weight – (a)   × Gradient.

The main layer in CNN that detects local features such as edges is the (a)   layer.

Reducing the spatial dimension of feature maps in CNN is usually done using (a)   layers.

CNNs use shared weights called (a)   to extract features from input images.

The technique of adding zeros around the input to preserve dimensions in CNN is called (a)   .

The CNN architecture that introduced skip connections is called (a)   .

The (a)   function is the most commonly used activation function in CNN due to reduced vanishing gradient issues.

For object detection tasks, architectures such as (a)   and Faster R-CNN are widely used.

CNN models trained on large datasets can be reused for new tasks using (a)   learning.

In CNN, the number of learnable parameters is reduced by using (a)   weights across receptive fields.

A CNN designed for biomedical image segmentation that uses skip connections between encoder and decoder is called (a)   .